Detection and segmentation of faces using binary partition trees

In this paper we improve the face detection and segmentation technique proposed in [1]. In order to obtain the shape of the face, we use a region based approach and find the face as a set of regions from a generic segmentation. The original image is segmented and a partition tree is created by merging regions from this partition. Facial descriptors and a similarity measure to faces are computed for each node. The analysis is performed using information from the regions represented by the node and also information from neighboring regions. The new method overcomes the rigidity of the tree structure and allows the extraction of new facial regions that are not represented as nodes in the tree. A search algorithm selects the nodes associated to faces. The use of information from neighboring regions significantly improves the performance of the algorithm and avoids the postprocessing step used in our previous work to completely extract the facial regions. Eje: Programación de imágenes Red de Universidades con Carreras en Informática (RedUNCI)